Department : Meteorological Engineering Programme : Atmospheric Sciences
ĠSTANBUL TECHNICAL UNIVERSITY INSTITUTE OF SCIENCE AND TECHNOLOGY
M.Sc. Thesis by Abdullah KAHRAMAN
SEPTEMBER 2009
ĠSTANBUL TECHNICAL UNIVERSITY INSTITUTE OF SCIENCE AND TECHNOLOGY
M.Sc. Thesis by Abdullah KAHRAMAN
511061102
Date of submission : 28 August 2009 Date of defence examination: 1 September 2009
Supervisor (Chairman) : Prof. Dr. Mikdat KADIOĞLU (ITU) Members of the Examining Committee : Assoc. Prof. Dr. Yurdanur ÜNAL (ITU))
Assoc. Prof. Dr. Ömer Lütfi ġEN (ITU)
SEPTEMBER 2009
EYLÜL 2009
ĠSTANBUL TEKNĠK ÜNĠVERSĠTESĠ FEN BĠLĠMLERĠ ENSTĠTÜSÜ
YÜKSEK LĠSANS TEZĠ Abdullah KAHRAMAN
511061102
Tezin Enstitüye Verildiği Tarih : 28 Ağustos 2009 Tezin Savunulduğu Tarih : 1 Eylül 2009
Tez DanıĢmanı : Prof. Dr. Mikdat KADIOĞLU (ĠTÜ) Diğer Jüri Üyeleri : Doç. Dr. Yurdanur ÜNAL (ĠTÜ)
Doç. Dr. Ömer Lütfi ġEN (ĠTÜ) ġĠDDETLĠ BĠR KONVEKSĠYON OLAYININ ORTA ÖLÇEKLĠ ANALĠZĠ
FOREWORD I would like to thank my advisor Prof. Dr. Mikdat Kadıoğlu for his guidance and patience during the preparation of this thesis. I would like to express my deep appreciation and special thanks for Prof. Dr. David Schultz for his contribution to this work. I also thank to Şeyda Tanrıöver, Daniele Mastrangelo, Deniz Ural, Serkan Eminoğlu, Kari Niemelä, Gökay Bıyık and Tayfun Dalkılıç for their helps, as well as Prof. Dr. Zerefşan Kaymaz, who encouraged me to study at University of Helsinki. All computations in this study are performed at the Meteorological Modelling Laboratory at ITU.
August 2009 Abdullah Kahraman
TABLE OF CONTENTS
Page ABBREVIATIONS ... Hata! Yer işareti tanımlanmamış.x LIST OF TABLES ... Hata! Yer işareti tanımlanmamış. LIST OF FIGURES ... xiHata! Yer işareti tanımlanmamış.
SUMMARY ... xviiii
ÖZET...xix
1. INTRODUCTION ...1
2. SEVERE CONVECTION AND RECENT STUDIES IN EUROPE ...3
2.1 Convection and Deep Moist Convection (DMC) ... 3
2.2 Recent Studies About Severe Weather in Europe ... 5
3. CASE STUDY OF A SEVERE CONVECTION EVENT IN TURKEY ...7
3.1 Objective ... 7
3.2 Methodology ... 7
3.2.1 Mesoscale modelling with WRF-ARW ... 7
3.3 Description of Study Area ...10
3.4 Data ...11
3.5 Sensitivity Analysis ...15
3.6. Synoptic Analysis ...45
3.7. Mesoscale Analysis ...69
4. CONCLUSION AND RECOMMENDATIONS ... 81
REFERENCES ... 83
APPENDICES ... 89
ABBREVIATIONS
ARW : Advanced Research WRF
CAPE : Convective Available Potential Energy CCL : Cumulative Condensation Level CT : Cross Total Index
DMC : Deep Moist Convection EL : Equilibrium Level
LCL : Lifting Condensation Level LI : Lifted Index
LFC : Level of Free Convection LSM : Land Surface Model
MCC : Mesoscale Convective Complex MCS : Mesoscale Convective System NWP : Numerical Weater Prediction RH : Relative Humidity
SREH : Storm Relative Helicity TT : Total Total Index VT : Vertical Total Index
WPS : WRF Preprocessing System
LIST OF TABLES
Page Table 3.1: USGS 24-category Land Use Categories ... 12 Table 3.2: 16-category Soil Categories ... 12 Table 3.3 : 28 parameters from NCEP Parameter Code Table 1 included in NCEP
final operational analysis dataset. ... 13 Table 3.4 : 32 parameters from NCEP Parameter Code Table 2 included in NCEP
final operational analysis dataset. ... 14 Table 3.5 : Sensitivity Analysis Elements. ... 16 Table 3.6: RMSE and BIAS statistics for the second domain of 15th experiment. .. 21 Table 3.7: RMSE and BIAS statistics for the second domain of 16th experiment. .. 22 Table 3.8: RMSE and BIAS statistics for the second domain of 17th experiment. .. 22 Table 3.9: RMSE and BIAS statistics for the second domain of 24th experiment. .. 22 Table 3.10: RMSE and BIAS statistics for the second domain of 27th experiment.. 23 Table 3.11: RMSE and BIAS statistics for the second domain of 18th experiment.. 23 Table 3.12: RMSE and BIAS statistics for the second domain of 19th experiment.. 24 Table 3.13: RMSE and BIAS statistics for the second domain of 20th experiment.. 24 Table 3.14: RMSE and BIAS statistics for the second domain of 21st experiment. . 24 Table 3.15: RMSE and BIAS statistics for the second domain of 22nd experiment. 25 Table 3.16: RMSE and BIAS statistics for the second domain of 23rd experiment. 25 Table 3.17: RMSE and BIAS statistics for the second domain of 30th experiment.. 26
LIST OF FIGURES
Page Figure 2.1 : Classic thunderstorm cell schematics from Byers and Braham (1949)
showing the (a) the cumulus stage, (b) the mature stage, and (c) the
dissipating stage. ... 4
Figure 3.1: Flow Chart for WRF-ARW. ... 8
Figure 3.2: ARW sigma coordinate. ... 9
Figure 3.3: Map of the area studied. ...10
Figure 3.4: Land use category for 1st domain with 24 km horizontal resolution. ....11
Figure 3.5: Domains of the sensitivity analysis test elements 1 and 2.. ...15
Figure 3.6: 1500 m relative vorticity plots of the 2nd domain of 4th experiment. ...18
Figure 3.7: Third domain’s terrain height above mean sea level, used at 4th, 5th and 6th experiments. The mountainous zone at the northwest is significant. ...19
Figure 3.8: 1500 m relative vorticity plots of the third domain of experiment 12 for t+3, t+15, t+40 and t+44. See the topography at western lateral boundary creating unnatural waves in all plots. ...20
Figure 3.9: Mean sea level pressure difference of the forecast and the observation at 12z on 15.08.2004, according to the sensitivity test elements. ...27
Figure 3.10: 850 hPa level equivalent potential temperature difference of the forecast and the observation at 12z on 15.08.2004, according to the sensitivity test elements. ...31
Figure 3.11 : Lowest sigma level temperature difference of the forecast and the observation at 12z on 15.08.2004, according to the sensitivity test element 30. ...35
Figure 3.12 : 1500 m relative vorticity difference of the forecast and the observation at 12z on 15.08.2004, according to the sensitivity test element 30. ...36
Figure 3.13 : 850 hPa level temperature difference of the forecast and the observation on 12z 15.08.2004, according to the sensitivity test 30. ..36
Figure 3.14 : 700 hPa level relative humidity difference of the forecast and the observation at 12z on 15.08.2004, according to the sensitivity test element 30. ...37
Figure 3.15 : 500 hPa relative vorticity difference of the forecast and the observation at 12z on 15.08.2004, according to the sensitivity test element 30. ...37
Figure 3.16 : 500 hPa temperature difference of the forecast and the observation at 12z on 15.08.2004, according to the sensitivity test element 30. ...38
Figure 3.17 : 925 hPa equivalent potential temperature difference of the forecast and the observation at 12z on 15.08.2004, according to the sensitivity test element 30. ...39
Figure 3.18 : CAPE difference of the forecast and the observation at 12z on 15.08.2004, according to the sensitivity test element 30. ...39
Figure 3.19 : 14.08.2004 00UTC 500 hPa chart...44
Figure 3.20 : 14.08.2004 00UTC 500 hPa relative vorticity. ...45
Figure 3.22 : 14.08.2004 00UTC 700 hPa relative humidity and wind. ... 47
Figure 3.23 : 14.08.2004 00UTC 850 hPa chart. ... 48
Figure 3.24 : 14.08.2004 00UTC 850 hPa equivalent potential temperature. ... 49
Figure 3.25 : 14.08.2004 00UTC 925 hPa equivalent potential temperature. ... 50
Figure 3.26 : 14.08.2004 00UTC 925 hPa relative vorticity. ... 51
Figure 3.27 : 14.08.2004 00UTC surface weather chart, analysed by DWD. ... 52
Figure 3.28 : 14.08.2004 01UTC lowest level temperature and 10 m wind. ... 53
Figure 3.29: 14.08.2004 00UTC streamlines at 10 m. ... 54
Figure 3.30 : 500 hPa plots between 14.08.2004 06UTC to 15.08.2004 12UTC, with 6 hours interval. ... 55
Figure 3.31 : 500 hPa relative vorticity plots between 14.08.2004 06UTC to 15.08.2004 12UTC, with 6 hours interval. ... 56
Figure 3.32 : 300 hPa streamlines and jets between 14.08.2004 06UTC to 15.08.2004 12UTC, with 6 hours interval. ... 58
Figure 3.33 : 700 hPa RH, horizontal wind and vertical velocity plots between 14.08.2004 06UTC to 15.08.2004 12UTC, with 6 hours interval. ... 59
Figure 3.34 : 850 hPa charts between 14.08.2004 06UTC to 15.08.2004 12UTC, with 6 hours interval. ... 61
Figure 3.35 : 850 hPa equivalent potential temperature plots between 14.08.2004 06UTC to 15.08.2004 12UTC, with 6 hours interval. ... 62
Figure 3.36 : 925 hPa equivalent potential temperature plots between 14.08.2004 06UTC to 15.08.2004 12UTC, with 6 hours interval. ... 64
Figure 3.37 : 925 hPa relative vorticity plots between 14.08.2004 06UTC to 15.08.2004 12UTC, with 6 hours interval. ... 65
Figure 3.38 : Surface analysis charts from 06UTC on 14.08.2004 to 12UTC on 15.08.2004, issued by DWD. ... 66
Figure 3.39 : Domain 1, CAPE & SREH at 05UTC, 15.08.2004. ... 69
Figure 3.40 : Domain 3, CAPE & SREH at 05UTC, 15.08.2004. ... 70
Figure 3.41 : Skew-T log-p diagram and hodograph of Istanbul for 05UTC on 15.08.2004. ... 71
Figure 3.42 : Streamlines at 10 m above surface, for 04UTC and 05UTC, 15.08.2004. ... 72
Figure 3.43 : 10 m wind barbs and ground level temperature, for 04UTC and 05UTC, 15.08.2004. ... 72
Figure 3.44 : 850 hPa wind and temperature, for 04UTC, 05UTC and 06 UTC, 15.08.2004. ... 73
Figure 3.45 : 700 hPa humidity, wind and vertcal velocity for 02UTC, 03UTC, 04UTC, 05UTC, 06 UTC and 07UTC on 15.08.2004. ... 74
Figure 3.46 : 500 hPa geopotential height, temperature and wind for 04UTC, 05UTC and 06 UTC on 15.08.2004. ... 75
Figure 3.47 : 10 m streamlines for 05 UTC on 15.08.2004. ... 76
Figure 3.48 : Vertical cross sections from SW to NE of the domain 3. From 00UTC to 05UTC on 15.08.2004. ... 77
Figure 3.49 : Composite chart of the case at 12UTC 15.08.2004, on METEOSAT-7 WV image. Blue line is the 300 hPa jet, magenta line is the low-level jet, red crosses are the maximum vorticity areas, black line is the cold front on surface and dashed green line is the significant isotherm (-12 C) at 500 hPa. ... 79
Figure A.1 : Output of Run 30: Synoptic conditions that favor the convection. ... 91
Figure A.3 : NCEP Reanalysis plots for 14.08.2004 and 15.08.2004. ... 105 Figure A.4: Total Hourly Precipitation on 06z, 15.08.2004, according to the
sensitivity test elements. ... 109 Figure A.5 : TRMM rainfall estimation for 15.08.2004. ... 113
MESOSCALE ANALYSIS OF A SEVERE CONVECTION EVENT SUMMARY
Severe weather events are usually related with deep moist convection (DMC), and must be studied with mesoscale processes that favor them. Cloud resolving models are used to analyse severe convection, as well as high resolution observational networks. In this study, a severe convection case of 15 August 2004 in Marmara Region is analysed with WRF-ARW atmospheric model results, from synoptic scale to mesoscale, initialized with GFS data.
A broad investigation on the sensitivity of the model is tested before the simulation. Three domains with horizontal resolutions of 24 km (time step: 72 seconds), 8 km and 2.67 km with 45 vertical levels are used, creating the boundary conditions by two-way nesting procedure. After validation with observational data, cumulus scheme is chosen as Kain-Fritsch, WRF single moment 3-class-scheme is used for microphysics, NOAH land surface model is used with 4 layers, Monin-Obukhov with Zilitinkevich thermal roughness length scheme is used for surface physics, Mellor-Yamada-Janjic scheme is used for planetary boundary layer, and full diffusion dynamic option is used.
The model results are verified with traditional observations, as well as remote sensing data. The convection was very well forecasted by the model in synoptic scale. However, location and time of the tornadic supercell was not perfectly predicted, though success in creating the DMC cell with its severe features was satisfying.
The reasons of the severe convection are analysed with an ingredients-based methodology, namely the conditional instability, existence of LFC, and a lifting mechanism to make the parcel reach the LFC. In this case, synoptic and mesoscale effects played a role together to create a tornadic supercell. Conditional instability was existing with moderate CAPE values, and the moisture content was high enough due to the Black Sea. Frontal surface was a triggering factor for the convection. There was a convergence on the surface due to low level winds, favoring the updraft. Strong vertical wind shear was also a key factor for the severity of the supercell storm.
BĠR ġĠDDETLĠ KONVEKSĠYON OLAYININ ORTA ÖLÇEKLĠ ANALĠZĠ ÖZET
Şiddetli hava olayları genel olarak derin nem konveksiyonu ile ilişkili olduklarından, orta ölçekli yaklaşımlarla incelenmelidirler. Şiddetli konveksiyon analizi için yüksek çözünürlüklü atmosfer modelleri kullanılmakta, yine radar, uydu vb yüksek çözünürlüklü gözlem verisi ile çalışmalar yapılmaktadır. Bu çalışmada 15 Ağustos 2004’te Marmara Bölgesi’nde meydana gelen bir şiddetli konveksiyon hadisesi ele alınmış, NCEP GFS verisi kullanılarak WRF-ARW atmosfer modeli ile elde edilen sonuçlar sinoptik ve orta ölçekte değerlendirilmiştir.
Çalışmadan önce modelleme için büyük çapta bir hassaslık analizi gerçekleştirilmiştir. 24 km (zaman adımı: 72 saniye), 8 km ve 2.67 km’lik yatay çözünürlüğe sahip üç domain, 45 düşey seviye ile çift-yön-nesting yöntemiyle çalıştırılmıştır. Yapılan verifikasyon sonucunda, Kain-Fritsch konvektif parametrizasyonu, WRF single moment 3-sınıf mikrofizik yaklaşımı, 4 seviyeli NOAH yer-yüzey modeli, Monin-Obukhov ve Zilitinkevich yüzey parametrizasyonu, Mellor-Yamada-Janjic atmosferik sınır tabaka modeli, ve tam difüzyon dinamik opsiyonu başarılı bulunmuştur.
Model sonuçları geleneksel gözlemlerin yanısıra uydu verileri ile de karşılaştırılmış, ve sinoptik ölçekte konveksiyonun gayet başarılı bir şekilde simüle edilebildiği görülmüştür. Öte yandan, tornado üreten super-hücreli fırtınanın tahmininde konum ve zaman olarak sapma gerçekleşmiş, ancak hücrenin pek çok özellikleriyle modellenebilmesi de tatmin edici bulunmuştur.
Şiddetli konveksiyonun nedenleri içerik-bazlı metodoloji ile incelenmiştir. Bu metodolojiye göre gerekli olan üç içerik, koşullu kararsızlık, LFC’nin mevcudiyeti, ve parseli LFC’ye taşıyacak bir kaldırma mekanizmasıdır. Konveksiyonun şiddetinde sinoptik ve orta ölçekli süreçlerin birlikte rol oynadığı tespit edilmiştir. Bölgede orta seviyedeki CAPE değerleri ile koşullu kararsızlık ve Karadeniz’den kaynaklanan yüksek nem mevcuttur. Cephesel yüzey, konveksiyonu tetikleyen bir mekanizma olmuştur. Alçak seviye rüzgarlarının neden olduğu yüzey konverjansı yükselişi desteklemiş, yüksek düşey rüzgar şiri de süper-hücreli fırtınanın şiddetinde anahtar bir rol oynamıştır.
1. INTRODUCTION
Although there are not many studies done in region, convective storms can cause severe weather and significant damage in Turkey. On 15.08.2004, parts of Marmara Region and western Black Sea were effected by severe convection that produced a tornadic supercell near Yalova. The F1 tornado caused no causalities since it was not over the urban area, but the farms and fields of 455 people in 10 villages were substiantally damaged. Furthermore, high precipitation including hail with 3-4 cm diameter occured over the area, according to the newspapers. Meteorologists have forecasted severe convection and related phenomena more than 3 days before the event and issued early warnings on time. However, no spesific warnings about the small supercell or tornado could be done due to lack of a working meteorological radar and unpredictability.
One of the objectives of this study is to create a scientific awareness on severe weather occurancy over Turkey. Due to the fact that severe convection usually occurs over smaller scales of areas than the conventional observational network, voluntary reporters are essential for a database of such phenomena. In Turkey, this consciousness has yet started to appear. Since there is not reliable data about the severe convection storms in Turkey, there is also not any scientific study about the climatology of such storms done. In second chapter, a background over the area and related studies done in Europe are examined.
The main purpose of the thesis is applying a mesoscale modelling approach to a severe convection case in Turkey. This study can help improving the predictability of similar future events. In third chapter, the case is studied with a state-of-the-art mesoscale model, WRF-ARW, after a broad sensitivity analysis. Using the model results, an ingredients-based methodology is used to understand the phenomena, and reasons of severe convection are discussed as synoptic and mesoscale features. Fourth chapter states the conclusion and recommendations.
2. SEVERE CONVECTION AND RECENT STUDIES IN EUROPE
2.1 Convection and Deep Moist Convection (DMC)
Convection is the transport of some property by fluid movement (Doswell, 2001, p. 1). This term is generally used for expressing heat transport, in addition to the other two processes conduction and radiation.
Atmospheric convection is the heat transport in the atmosphere by the vertical component of the flow associated with buoyancy. Advection, either vertical or horizontal, refers to the transport of heat (or another property) by the nonbuoyant flow. Buoyancy is defined as follows:
' ' T T T g B 1.1
where g is the gravitational acceleration, T is the temperature of an air parcel, and T’ is the environmental temperature. When the temperature of the air parcel is higher than that of the environment, buoyancy is positive, where lower parcel temperature makes the buoyancy negative.
Integration of buoyancy from the Level of Free Convection (LFC) to Equilibrium Level (EL) gives the Convective Available Potential Energy (CAPE), which is frequently used in analysing and forecasting severe convection.
dz T T T g CAPE zEL zLFC
' ' 2.2Sometimes, hazardous weather can be associated with convection without thunder. That is why the term “Deep, Moist Convection” (DMC) is preferred instead of the term “thunderstorm”, by some scientists.
Figure 2.1 : Classic thunderstorm cell schematics from Byers and Braham (1949) showing the (a) the cumulus stage, (b) the mature stage, and (c) the dissipating stage.
The origin of modern concepts of DMC is the famous Thunderstorm Project and Byers and Braham’s report (1949). Other projects followed the path of this study, like the mesonetwork used by Fujita (1955), the NSSL mesonetworks analyzed by Barnes (1978), the National Hail Research Experiment described by Foote and Knight (1979) and the Florida Area Cumulus Experiment described by Barnston et al. (1983). Doswell states this as “a key concept grew out of this project: the thunderstorm cell” (2001). As depicted in Byers and Braham (1949), the cell is the basic organizational structure of all thunderstorms (Fig. 1.1.a, Fig. 1.1.b, Fig. 1.1.c) and this notion became the fundamental paradigm for deep moist convection (Doswell, 2001). Browning (1964) created the idea of supercell, using the thunderstorm cell paradigm as a basis for his ideas. This concept was also used for a taxonomy of severe hailstorms developed by Marwitz (1972a,b,c). Warner 1970, 1972; Simpson 1971, 1972; and Newton 1963 worked on how the entrainment was occurring in thunderstorms. Stommel’s (1947), Scorer and Ludlam’s (1953), Squires and Turner’s (1962) ideas on entrainment of environmental air to be occuring along the lateral boundaries of the cloud was criticized by them, like penetrative downdrafts within the cloud might have been contributing substantially to entrainment. Levine’s (1959) bubble theory for entrainment was rediscovered by Blyth (1993) with a modification, proposing that the major form of entrainment for convection is via the "toroidal" circulation like a smoke ring, associated with the
Doswell states that our perceptions of convective structure are highly dependent on the observing system being employed. Classifications based on radar are not necessarily equivalent to those developed from satellite imagery, and so on. The most intense severe convective weather is associated with organized DMC (2001). The term “organized” refers to squall lines, mesoscale convective systems (MCS) and supercells. The details of these DMC structures will not be reviewed here.
2.2 Recent Studies About Severe Weather in Europe
There are many studies done about severe storms and tornadoes not only in USA, but also in whole world. Here, some of the recent significant studies in Europe will be cited.
Tornado climatologies of Austria and Portugal can be some examples for creating one for Turkey. Holzer performed a tornado climatology of Austria in 2001, using a database of 89 tornadoes, one landspout and 6 waterspouts. Leitao (2003) also made a tornado climatology of Portugal, using 30 tornadoes that occurred from 1936 to 2002.
Garcia-Herrera et al (2005) studied on the characterization of mesoscale convective systems over Spain with a 3-year database. Rigo and Llasat (2007) worked on a climatology of mesoscale convective systems in the northeast of the Iberian Peninsula, on the basis of meteorological radar observations, using the period of 1996 to 2000. Sanches et al (2003) presented an analysis based on a classification of METEOSAT images for hail events in the Ebro Valley over Northeastern Spain. A logistic regressive analysis has been applied in the study and the results showed the difficulties encountered in forecasting the formation of MCSs on the basis of preconvective variables.
In addition to general studies, there are also many case studies performed. Lopez (2007) performed a case study for a derecho over Catalonia using a high resolution observation network and radar data. Takemi (2007) studied on the sensitivity of squall-line intensity to environmental static stability under various shear and moisture conditions. Putsay and friends (2008) made a case study of mesoscale convective systems over Hungary on 29 June 2006 with satellite, radar and lightning
data. Furthermore, Kaspar et al (2008) studied a severe storm in Bavaria, the Czech Republic and Poland on 12–13 July 1984. They performed a statistic- and model-based analysis. Aran et al (2008) made a synoptic and mesoscale analysis of an F2 tornado event over Catalonia, and Kaltenböck (2004) studied a mesoscale analysis of a case study of a MCS over northeast of the Alps. These high resolution observation network based studies are maybe the most objective way to examine such phenomena.
The most populer way of studying severe weather is numerical modelling. Modern atmospheric models have the capability of simulating mesoscale meteorological features with high accuracy, when used with a good initial condition and proper physics / dynamics packages. Some of the studies done with models are given here. Knupp et al (1998) described a mini-supercell storm over northern Alabama through a combined observational analysis and numerical modelling study. RAMS was used as the modelling tool. Martin et al (2006)’s case study on a heavy precipitation event was performed with MM5 simulations. Ortega and friends (2008) worked on a sensitivity study of two severe storm cases in the Southeastern Andes using MM5. Keil and Hagen (2000) evaluated high resolution simulations with radar data, using two models; Deutschland Model and Canadian MC2 Research model. Krichak and friends (2000) studied a severe convection case of 02.11.1994 in the southeastern Mediterranean using MM5. Horvath et al (2008) wrote a paper on numerical modeling of severe convective storms occurring in the Carpathian Basin. They examined two case studies, one pre-frontal squall line and one frontal convective line, using MM5. Garcia-Ortega et al (2007) made a sensitivity study of a severe hailstorm in northeasterm Spain using MM5. Bechtold and Bazile (2001) worked on a case study of a flash flood over southern France using ALADIN with ARPEGE input. The methodology such studies follow is to simulate the atmospheric conditions accurately, verified by observations, and perform synoptic to mesoscale analysis subjectively. The case study in third chapter also follows this path.
3. CASE STUDY OF A SEVERE CONVECTION EVENT IN TURKEY
3.1 Objective
Main objective of this case study is to analyse a severe convection event with respect to its mesoscale properties, to define a severe weather phase on the region. As the number of such studies will increase, operational forecasting of severe weather events and early warnings will be able to be performed accurately.
3.2 Methodology
An ingredients-based methodology is used within this study. Doswell’s three ingredients for deep moist convection are basicly known as the conditional instability, a moisture source and a lifting mechanism. The question is, how these three ingredients occured and came together at our case. As the physical reasons of these ingredients will be able to be explained, the analysis will be done.
In order to see how the three ingredients come together, we must have a mesoscale analysis of the meteorological conditions of the case. Mostly, meteorological stations are not dense enough with respect to horizontal resolution, to analyse a convection case which requires a fine scale observation network in terms of space and time. To create a satisfactory high resolution analysis, mesoscale models can be used to dynamically downscale coarse resolution data. In this study, this approach is used to investigate the mesoscale analysis of the severe weather case.
3.2.1 Mesoscale modelling with WRF-ARW
The Weather Research and Forecasting (WRF) model is known as a replacement of MM5, a numerical weather prediction (NWP) and atmospheric simulation system designed for both research and operational applications. The development of WRF has been a multi-agency effort to build a next-generation mesoscale forecast model
and data assimilation system to advance the understanding and prediction of mesoscale weather and accelerate the transfer of research advances into operations. The WRF effort has been a collaborative one among the National Center for Atmospheric Research’s (NCAR) Mesoscale and Microscale Meteorology (MMM) Division, the National Oceanic and Atmospheric Administration’s (NOAA) National Centers for Environmental Prediction (NCEP) and Earth System Research Laboratory (ESRL), the Department of Defense’s Air ForceWeather Agency (AFWA) and Naval Research Laboratory (NRL), the Center for Analysis and Prediction of Storms (CAPS) at the University of Oklahoma, and the Federal Aviation Administration (FAA), with the participation of university scientists in USA (Skamarock et al., 2008).
WRF has a wide physics and dynamics options, as well as a data assimilation and chemistry modelling tool.
The model uses terrain-following, hydrostatic-pressure vertical coordinate (Figure 3.2) with the top of the model being a constant pressure surface. The horizontal grid WRF uses is the Arakawa-C grid. The time integration scheme in the model uses the third-order Runge-Kutta scheme, and the spatial discretization employs 2nd to 6th order schemes. Lamber Comformal, Polar Stereographic and Mercator are the three projections WRF supports.
Figure 3.2: ARW sigma coordinate
To make a real data forecast with WRF-ARW, three steps are followed. First is the preprocessing, second is the model run, and the last is the visualisaton part. In figure 3.1, flow chart of WRF-ARW can be seen.
Preprocessing system of WRF has three stages; geogrid, ungrib and metgrid. Geogrid is a program which creates the domain according to the WPS namelist and TBL files. The static data including topography, terrains, land use etc are processed at this stage per domain. Ungrib performs degribbing the gribbed atmospheric variables data. Finally, metgrid puts the output of the geogrid and ungrib together on new netcdf files, ready for the real.exe.
After the completion of preprocessing, met_em files for the whole time of the simulation must be ready for the ARW model. These files are used as input to real.exe, which is a program interpolating the fields to the vertical levels defined by the namelist.input file of the WRF. After the real.exe process is succesfully completed, the model solver starts its calculation and performs the simulation.
Output of the model can be visualised with tools like RIP, NCL, GrADS, PyNGL, Vis5D, VAPOR, etc.
3.3 Description of Study Area
The area of the case study is The Marmara Region of Turkey, which is the most populized and industrialized part of the country. Including the historical Istanbul, where is also the heart of the business and culture in Turkey, surrounded by Black Sea, Eagean Sea, Balkans, Anatolia, and including an inland sea –Marmara-, the area has been always important throughout the ages.
With its mid-latitute geography and complex topography, Eastern Mediterranean is an area which has very interesting climatological features. Marmara Region lays down on the path of polar fronts, even sometimes arctic fronts visit the area. In addition to such frontal phases, it also receives the Asor and Siberia Highs’ effects, with a seasonal variability. Dry air from high mountains of Balkans and Anatolia, as well as moist air from Marmara, Black Sea and Eagean have different aspects in weather conditions in whole region.
Figure 3.3 shows a map of the area studied.
3.4 Data
The data used in the model is of two types: The static data and the variable data. The static data includes the geographical features like terrain height, land use, soil categories, provided by the WRF model. Four different resolutions can be chosen according to the resolution interest for the studies. The resolutions are 10’, 5’, 2’ and 30”. Within this study, 10’ data is used for the first domain (24 km horizontal resolution), 5’ data is used for the second domain (8 km horizontal resolution) and 30” data is used for the third domain (2.7 km horizontal resolution). All static fields are used as an input for geogrid, which is a WPS program interpolating and locating the data into the selected domain’s grid points. Table 3.1 is the table of the land use, and Table 3.2 is of the soil categories defined within the static data. Figure 3.4 shows how the land use is described in the first domain.
Table 3.1: USGS 24-category Land Use Categories Land Use Category Land Use Description
1 Urban and Built-up Land
2 Dryland Cropland and Pasture
3 Irrigated Cropland and Pasture
4 Mixed Dryland/Irrigated Cropland and Pasture
5 Cropland/Grassland Mosaic 6 Cropland/Woodland Mosaic 7 Grassland 8 Shrubland 9 Mixed Shrubland/Grassland 10 Savanna
11 Deciduous Broadleaf Forest
12 Deciduous Needleleaf Forest
13 Evergreen Broadleaf 14 Evergreen Needleleaf 15 Mixed Forest 16 Water Bodies 17 Herbaceous Wetland 18 Wooden Wetland
19 Barren or Sparsely Vegetated
20 Herbaceous Tundra
21 Wooded Tundra
22 Mixed Tundra
23 Bare Ground Tundra
24 Snow or Ice
Table 3.2: 16-category Soil Categories Soil Category Soil Description
1 Sand 2 Loamy Sand 3 Sandy Loam 4 Silt Loam 5 Silt 6 Loam
7 Sandy Clay Loam
8 Silty Clay Loam
9 Clay Loam 10 Sandy Clay 11 Silty Clay 12 Clay 13 Organic Material 14 Water 15 Bedrock 16 Other (land-ice)
The variable data includes the variables of the atmosphere, oceans and lands. NCEP final operational analysis data is used as the variable-input to the model. The data are on 1.0x1.0 degree grids continuously at every six hours. This product is from the Global Forecast System (GFS) that is operationally run four times a day in near-real time at NCEP. The analyses are available on the surface, at 26 mandatory (and other pressure) levels from 1000mb to 10mb, in the surface boundary layer and at some sigma layers, the tropopause and a few others. Data is in GRIB format, which is converted into another raw format by the ungrib tool inside the WPS. Tables 3.3 and 3.4 show the variables included in this dataset.
Table 3.3 : 28 parameters from NCEP Parameter Code Table 1 included in NCEP final operational analysis dataset.
Parameter
Code Name Short Description Units
1 PRES Pressure Pa
2 PRMSL Pressure reduced to MSL Pa
7 HGT Geopotential height gpm
10 TOZNE Total ozone Dobson
11 TMP Temperature K
13 POT Potential temperature K
27 GPA Geopotential height anomaly gpm
33 U GRD u-component of wind m s-1
34 V GRD v-component of wind m s-1
39 V VEL Pressure vertical velocity Pa s-1
41 ABS V Absolute vorticity s-1
51 SPF H Specific humidity kg kg-1
52 R H Relative humidity %
54 P WAT Precipitable water kg m-2
65 WEASD Water equivalent of accumulated snow
depth kg m
-2
71 T CDC Total cloud cover %
81 LAND Land cover fraction
91 ICE C Ice concentration fraction
131 LFT X Surface lifted index K
132 4LFTX Best (4 layer) lifted index K
136 VW SH Vertical speed shear s-1
144 SOILW Volumetric soil moisture content fraction
154 O3MR Ozone mixing ratio kg kg-1
156 CIN Convective inhibition J kg-1
157 CAPE Convective available potential energy J kg-1
221 HPBL Planetary boundary layer height m
222 5WAVH 5-wave geopotential height gpm
Table 3.4 : 32 parameters from NCEP Parameter Code Table 2 included in NCEP final operational analysis dataset.
Parameter
Code Name Short Description Units
1 PRES Pressure Pa
2 PRMSL Pressure reduced to MSL Pa
7 HGT Geopotential height gpm
10 TOZNE Total ozone Dobson
11 TMP Temperature K
13 POT Potential temperature K
15 TMAX Maximum temperature K
16 TMIN Minimum temperature K
27 GPA Geopotential height anomaly gpm
33 UGRD u-component of wind m s-1
34 VGRD v-component of wind m s-1
39 VVEL Pressure vertical velocity Pa s-1
41 ABSV Absolute vorticity s-1
51 SPFH Specific humidity kg kg-1
52 RH Relative humidity %
54 PWAT Precipitable water kg m-2
65 WEASD Water equivalent of accumulated snow
depth kg m
-2
71 TCDC Total cloud cover %
76 C WAT Cloud water kg m-2
81 LAND Land cover fraction
91 ICEC Ice cover fraction
131 LFTX Surface lifted index K
132 4LFTX Best (4 layer) lifted index K
136 VSSH Vertical speed shear s-1
144 SOILW Volumetric soil moisture content fraction
153 CLWMR Cloud water mixing ratio kg kg-1
154 O3MR Ozone mixing ratio kg kg-1
156 CIN Convective inhibition J kg-1
157 CAPE Convective available potential energy J kg-1
221 HPBL Planetary boundary layer height m
222 5WAVH 5-wave geopotential height gpm
230 5WAVA 5-wave geopotential height anomaly
In addition to the model input data, satellite images of METEOSAT-7 supplied by Turkish State Meteorological Service and TRMM data retrieved from NASA are used for verification.
3.5 Sensitivity Analysis
In order to understand the sensitivity of the model to the lateral boundaries, physical parameterizations, dynamical approaches, horizontal resolution, number of vertical levels, time step and other parameters, several simulations have been performed before the experiment. Table 3.1 shows the elements of the tests and namelist variables.
The first aim of the tests was to determine the lateral boundaries for the simulation. Lateral boundaries of a selected domain has a very important role in simulation, because lateral boundary conditions depend on this. If the observations at the lateral boundaries do not represent the “real” weather conditions, then the regional model will create “unreal” patterns in whole domain. To adjust the best lateral boundaries, one must choose topographically homogeneous and uniform areas as the edges of the domain. This helps the possibility of the topography of the model domain to be more similar to that of the global model output or observation data used, as well as calculations of the relaxation zone to be more “natural”. That’s why three different domains are tested as the first experiments, with all other parameters being constant. Figure 3.5 shows the first two domains used in sensitivity analysis elements 1 and 2. The first domain was found to be too large for a mesoscale model run due to the planetary scale facts and also with respect to computational time. The final study is performed with a domain smaller than both these domains.
Figure 3.5: Domains of the sensitivity analysis test elements 1 and 2, named as (a) and (b). Horizontal resolutions of both domains are 24 km.
Table 3.5. Sensitivity Analysis Elements.
RUN domain cu mp radiation initalization lsm layer
time step 1 a 1 3 1 14.08, 00:00 1 5 120 2 b 1 3 1 14.08, 00:00 1 5 120 3 c 1 3 1 14.08, 00:00 1 5 120 4 b 1: 3: 5 1,1,0 3,3,3 1,1,1 14.08, 00:00 1,1,1 5,5,5 120 5 b 1: 3: 5 5,5,5 3,3,3 1,1,1 14.08, 00:00 1,1,1 5,5,5 120 6 b 1: 3: 5 2,2,2 3,3,3 1,1,1 14.08, 00:00 1,1,1 5,5,5 120 7 d 1 1 3 1 14.08, 00:00 1 5 120 8 d 1: 3: 5 1,1,0 3,3,3 1,1,1 14.08, 00:00 1,1,1 5,5,5 120 9 d 1: 3: 5 5,5,5 3,3,3 1,1,1 14.08, 00:00 1,1,1 5,5,5 120 10 d 1: 3: 5 2,2,2 3,3,3 1,1,1 14.08, 00:00 1,1,1 5,5,5 120 11 d 1: 3: 5 1,1,0 5,5,5 1,1,1 14.08, 00:00 1,1,1 5,5,5 120 12 d 1: 3: 5 1,1,0 5,5,5 99,99,99 14.08, 00:00 1,1,1 5,5,5 120 13 d 1: 3: 5 1,1,0 3,3,3 1,1,1 15.08, 00:00 1,1,1 5,5,5 120 14 d 1: 3: 5 5,5,5 3,3,3 1,1,1 15.08, 00:00 1,1,1 5,5,5 120 15 last 1: 3: 3 1,1,1 3,3,3 1,1,1 15.08, 00:00 1,1,1 5,5,5 120 16 last 1: 3: 3 5,5,5 3,3,3 1,1,1 15.08, 00:00 1,1,1 5,5,5 120 17 last 1: 3: 3 2,2,2 3,3,3 1,1,1 15.08, 00:00 1,1,1 5,5,5 120 18 last 1: 3: 3 1,1,1 3,3,3 1,1,1 15.08, 00:00 2,2,2 4 75 19 last 1: 3: 3 1,1,1 3,3,3 1,1,1 15.08, 00:00 1,1,1 5,5,5 120 20 last 1: 3: 3 1,1,1 5,5,5 1,1,1 15.08, 00:00 1,1,1 5,5,5 120 21 last 1: 3: 3 1,1,1 5,5,5 99,99,99 15.08, 00:00 1,1,1 5,5,5 120 22 last 1: 3: 3 1,1,1 3,3,3 1,1,1 15.08, 00:00 1,1,1 5,5,5 120 23 last 1: 3: 3 1,1,1 3,3,3 1,1,1 15.08, 00:00 1,1,1 5,5,5 120 24 last 1: 3: 3 3,3,3 3,3,3 1,1,1 15.08, 00:00 1,1,1 5,5,5 120 25 last 1: 3: 3 1,1,1 3,3,3 1,1,1 15.08, 00:00 1,1,1 5,5,5 120 26 last 1: 3: 3 1,1,1 3,3,3 1,1,1 15.08, 00:00 1,1,1 5,5,5 120 27 last 1: 3: 3 99,99,99 3,3,3 1,1,1 15.08, 00:00 1,1,1 5,5,5 120 28 last 1: 3: 3 1,1,1 3,3,3 1,1,1 15.08, 00:00 1,1,1 5,5,5 72 29 last 1: 3: 3 1,1,1 3,3,3 1,1,1 14.08, 12:00 1,1,1 5,5,5 120 30 last 1: 3: 3 1,1,1 3,3,3 1,1,1 15.08, 00:00 2,2,2 4 72 31 last 1: 3: 3 1,1,1 3,3,3 1,1,1 14.08, 18:00 1,1,1 5,5,5 120 32 last 1: 3: 3 1,1,1 3,3,3 1,1,1 15.08, 00:00 1,1,1 5,5,5 120 33 last 1: 3: 3 1,1,1 3,3,3 1,1,1 15.08, 00:00 2,2,2 4 72
Table 3.5. Sensitivity Analysis Elements (continued).
sfclay pbl co2tf rk_ord diff_6th_opt hydrostatic
vertical levels
spec_
bdy_width relax diff_opt data
1 1 2 non-h 27 4 1 GFS 1 1 2 non-h 27 4 1 GFS 1 1 2 non-h 27 4 1 GFS 1 1 2,2,2 non-h 27 4 1 GFS 1 1 2,2,2 non-h 27 4 1 GFS 1 1 2,2,2 non-h 27 4 1 GFS 1 1 2 non-h 27 4 1 GFS 1 1 2,2,2 non-h 27 4 1 GFS 1 1 2,2,2 non-h 27 4 1 GFS 1 1 2,2,2 non-h 27 4 1 GFS 1 1 2,2,2 non-h 27 4 1 GFS 1 1 2,2,2 non-h 27 4 1 GFS 1 1 2,2,2 non-h 27 4 1 GFS 1 1 2,2,2 non-h 27 4 1 GFS 1 1 2,2,2 non-h 27 4 1 GFS 1 1 2,2,2 non-h 27 4 1 GFS 1 1 2,2,2 non-h 27 4 1 GFS 1 1 2,2,2 non-h 27 4 1 GFS 2 2 2,2,2 non-h 27 4 1 GFS 1 1 2,2,2 non-h 27 4 1 GFS 1 1 1 2,2,2 non-h 27 4 1 GFS 1 1 3 1,1,1 non-h 27 4 1 GFS 1 1 2,2,2 hydrostatic 27 4 1 GFS 1 1 2,2,2 non-h 27 4 1 GFS 1 1 2,2,2 non-h 45,45,45 4 1 GFS 1 1 2,2,2 non-h 27 1 0 1 GFS 1 1 2,2,2 non-h 27 4 1 GFS 1 1 2,2,2 non-h 27 4 2 GFS 1 1 2,2,2 non-h 27 4 1 GFS 2 2 2,2,2 non-h 45,45,45 4 2 GFS 1 1 2,2,2 non-h 27 4 1 GFS 1 1 2,2,2 non-h 27 4 1 ECMWF 2 2 2,2,2 non-h 45,45,45 4 2 ECMWF
Experiments 4, 5 and 6 are performed with three domains using two-way nesting procedure with the domain b as the mother domain, to test three cumulus parameterization schemes (Kein Fritsch, Grell 3d ensemble, Betts-Miller-Janjic), holding all other parameters fixed. Second domain had a horizontal resolution of 8 km, where third had 1.6 km. Since the this is fine enough with respect to horizontal resolution, no cumulus parameterization is used at 4th experiment’s third domain. Grell-3d ensemble scheme is designed for a high resolution run according to the users guide of the model, so it is used in all domains of 5th exeriment. BMJ is also used in all three domains.
Figure 3.6: 1500 m relative vorticity plots of the second domain of 4th experiment. The boundaries of the third domain (surrounding the Marmara Region) create unnatural waves.
While these tests were performed to see the difference of precipitation patterns of three cumulus parameterization schemes, one other fact about the importance of domain choice appeared. Especially the relative vorticity plots showed clearly that there was something wrong at the edges of the very-fine-resolution third domain. Figure 3.6 shows the evolution of “unnatural” waves related to the lateral boundaries of this domain. Such patterns occurred also with the 5th and 6th experiments, so the reason of these patterns was not the cumulus parameterization scheme, but something different. Then the topography of the third domain is plotted, and the high mountains at the edges were thought to be the reason of these waves. Figure 3.7 shows the terrain height of the third domain used at 4th, 5th and 6th experiments.
Figure 3.7: Third domain’s terrain height above mean sea level, used at 4th, 5th and 6th experiments. The mountainous zone at the northwest is significant.
At seventh experiment, a new main domain is tested. 8th, 9th and 10th experiments are performed with this main domain and additional two nested domains to test the cumulus parameterization schemes again. Eta microphysics parameterization is used with 11th experiment, instead of WRF Single-Moment-3 class scheme. With 12nd experiment, GFDL (Eta operational scheme) radiation schemes (both for shortwave and longwave radations) are used with Eta microphysics. 13rd and 14th experiments were performed to see how different initial conditions effect the simulation results. Although this bunch of experiments (from 7th to 14th) showed how some main physical parameterizations effect the simulations, they have not been found accurate
Figure 3.8: 1500 m relative vorticity plots of the third domain of experiment 12 for t+3, t+15, t+40 and t+44. See the topography at western lateral boundary creating unnatural waves in all plots.
enough. The problem with the third domain’s lateral boundaries was not as bad as the 4th, 5th and 6th experiment’s one, but still remained at different points of the edges. Figure 3.8 shows some of the 1500 m level relative vorticity plots of the third domain of the 12nd run. At the western boundary, near the 170th grid point, a source of an “unnatural” wave is occuring. After these results, it is understood that the topography of the Balkans were too steep to be a boundary of a 1:5 aspect ratio nesting. That’s why, the aspect ratio between second and third domain is determined to be 1:3 like it is, between the first and second domain. All of the next experiments are performed with 1:3:3 aspect ratio domains, called as “last” domain choice. The details of the last domain can be found at the namelist.wps file, printed on page 40. 15th, 16th, 17th, 24th and 27th experiments are the cumulus parameterizations tests for the “last” domain choice. All five available schemes of the model could be tested with these runs. Results are compared with a script which calculates the RMSE and BIAS statistics of some main parameters of the whole output. Tables 3.6 to 3.10 show these statistics for cumulus parameterization tests.
18th experiment is performed with the Noah Land Surface Model and gave better results than the default 5-layer thermal diffusion as LSM, according to the RMSE and BIAS scores, which are shown at Table 3.7. Compared to the Table 3.6, it is obvious that forecasts of almost all parameteres are improved with this LSM.
Table 3.6: RMSE and BIAS statistics for the second domain of 15th experiment.
parameter RMSE BIAS
T -0.41980010 1.34151399 QVAPOR -0.00002452 0.00072632 QCLOUD 0.00000075 0.00001301 QRAIN 0.00000063 0.00001672 U -0.13463899 1.73248506 V -0.29448953 1.81327438 W -0.00246523 0.05883734 PH -16.09216309 70.96009827 PHB 0.02366149 127.23658752 MUB -0.28473541 180.04959106 U10 0.00305741 1.71484196 V10 -0.75020468 1.77986085
Table 3.7: RMSE and BIAS statistics for the second domain of 16th experiment.
parameter RMSE BIAS
T -0.44989517 1.35609746 QVAPOR -0.00001589 0.00073180 QCLOUD 0.00000130 0.00001892 QRAIN 0.00000018 0.00000434 U -0.13126636 1.72015059 V -0.30209583 1.80025303 W -0.00253718 0.05822026 PH -19.00615120 71.08341980 PHB 0.02366149 127.23658752 MUB -0.28473541 180.04959106 U10 -0.02382110 1.70702112 V10 -0.73529977 1.78133678
Table 3.8: RMSE and BIAS statistics for the second domain of 17th experiment.
parameter RMSE BIAS
T -0.42610377 1.35205960 QVAPOR -0.00002982 0.00073044 QCLOUD 0.00000088 0.00001564 QRAIN 0.00000077 0.00001742 U -0.13172835 1.77840269 V -0.29324630 1.85239935 W -0.00250439 0.05898664 PH -18.55657005 73.32422638 PHB 0.02366149 127.23658752 MUB -0.28473541 180.04959106 U10 0.00720553 1.72295427 V10 -0.74605560 1.78405702
Table 3.9: RMSE and BIAS statistics for the second domain of 24th experiment.
parameter RMSE BIAS
T -0.44126621 1.35820508 QVAPOR -0.00001782 0.00073187 QCLOUD 0.00000124 0.00001838 QRAIN 0.00000014 0.00000328 U -0.12867534 1.72293091 V -0.29637909 1.79107964 W -0.00261242 0.05949474 PH -17.13443375 71.49693298 PHB 0.02366149 127.23658752 MUB -0.28473541 180.04959106 U10 -0.00837070 1.70468199 V10 -0.71741563 1.76612532
Table 3.10: RMSE and BIAS statistics for the second domain of 27th experiment.
parameter RMSE BIAS
T -0.46403620 1.37078536 QVAPOR -0.00001840 0.00072533 QCLOUD 0.00000126 0.00001880 QRAIN 0.00000077 0.00002711 U -0.12704837 1.75677311 V -0.30051172 1.80252326 W -0.00254363 0.05933808 PH -19.02210426 70.72624969 PHB 0.02366149 127.23658752 MUB -0.28473541 180.04959106 U10 0.00684567 1.71931839 V10 -0.70963013 1.76146841
Table 3.11: RMSE and BIAS statistics for the second domain of 18th experiment.
parameter RMSE BIAS
T -0.43328232 1.32718968 QVAPOR 0.00000253 0.00065834 QCLOUD 0.00000074 0.00001284 QRAIN 0.00000058 0.00001456 U -0.10519053 1.71023738 V -0.23071042 1.79039860 W -0.00253149 0.05793478 PH -14.72518635 69.38451385 PHB 0.02366149 127.23658752 MUB -0.28473541 180.04959106 U10 0.07234187 1.65581787 V10 -0.60270309 1.68508255
Mellor-Yamada-Janjic scheme for planetary boundary layer (PBL) is tested with Monin-Obukhov with Zilitinkevich surface layer scheme option at 19th experiment. RMSE and BIAS scores of this test is shown at Table 3.12.
Experiment 20 and 21 have the Eta microphysics instead of WRF-Single-Moment-3 class scheme. 21 is performed with Eta radiation schemes for both shortwave and longwave. Statistics for these runs, which can be seen at Tables 3.13 and 3.14 indicate that these physical parameterizations are not better than the default ones.
Table 3.12: RMSE and BIAS statistics for the second domain of 19th experiment.
parameter RMSE BIAS
T -0.57406127 1.47866547 QVAPOR 0.00012830 0.00097297 QCLOUD 0.00000065 0.00001384 QRAIN 0.00000060 0.00001492 U -0.17349011 1.77619600 V -0.25964779 1.80933619 W -0.00302824 0.05015026 PH -23.34939194 76.33452606 PHB 0.02366149 127.23658752 MUB -0.28473541 180.04959106 U10 0.08273333 1.97083759 V10 -0.80356473 1.97025681
Table 3.13: RMSE and BIAS statistics for the second domain of 20th experiment.
parameter RMSE BIAS
T -0.34011883 1.33145094 QVAPOR -0.00004322 0.00073977 QCLOUD 0.00000057 0.00001398 QRAIN 0.00000013 0.00000546 U -0.13871635 1.74422026 V -0.28871560 1.83372617 W -0.00239768 0.05914829 PH -7.58681059 71.19757843 PHB 0.02366149 127.23658752 MUB -0.28473541 180.04959106 U10 0.03498381 1.74826205 V10 -0.72759897 1.78250623
Table 3.14: RMSE and BIAS statistics for the second domain of 21st experiment.
parameter RMSE BIAS
T -0.15759936 1.20540357 QVAPOR -0.00003912 0.00074395 QCLOUD 0.00000066 0.00001474 QRAIN 0.00000009 0.00000326 U -0.14336081 1.73450089 V -0.30870369 1.80103719 W -0.00223211 0.05793579 PH 9.97217274 70.98815155 PHB 0.02366149 127.23658752 MUB -0.28473541 180.04959106 U10 0.02296653 1.72466338 V10 -0.73885560 1.75841856
Instead of the default 2nd order, recomended 3rd order Runge-Kutta dynamics is used at experiment 22. diff_6th_opt parameter was also chosen as 1. However, as seen at Table 3.15, almost no improvement could be gained with this option, where the test was more expensive with respect to computation time.
23rd experiment was performed to see how accurrate the results would be, when the model is run with hydrostatic option. The simulation was a bit faster, but scores were not so much satisfactory. The RMSE and BIAS scores are calculated according to the second domain, which has a 8 km horizontal resolution, and almost all parameters are forecasted worse than non-hydrostatic run. So it is clear that with higher resolutions, this option will make the forecast just less accurate. Table 3.16 shows the statistics of hydrostatic run.
Table 3.15: RMSE and BIAS statistics for the second domain of 22nd experiment.
parameter RMSE BIAS
T -0.43502462 1.34549725 QVAPOR -0.00002176 0.00071989 QCLOUD 0.00000080 0.00001271 QRAIN 0.00000057 0.00001478 U -0.12689239 1.71436620 V -0.29390034 1.79741037 W -0.00256617 0.05229823 PH -18.11697960 70.93245697 PHB 0.02366149 127.23658752 MUB -0.28473541 180.04959106 U10 0.01191605 1.69555211 V10 -0.74891400 1.76443696
Table 3.16: RMSE and BIAS statistics for the second domain of 23rd experiment.
parameter RMSE BIAS
T -0.46731880 1.37395537 QVAPOR -0.00000522 0.00074576 QCLOUD 0.00000161 0.00002241 QRAIN 0.00000103 0.00002429 U -0.19031216 1.75898802 V -0.29910493 1.80942643 W 0.00054295 0.01059786 PH -26.68663979 73.68075562 PHB 0.02366149 127.23658752 MUB -0.28473541 180.04959106 U10 -0.15205720 1.72201800 V10 -0.77430010 1.74610722
25th experiment is performed to see the effect of number of vertical levels. Instead of default value 27, 45 levels are used with this run. Although the computational time takes longer, this brings out a better forecast especially with convection cases. The lateral boundaries can be used with or without relaxation at a regional model. Default WRF relaxation zone has 4 points. With 26th experiment, no relaxation is tested. No significant change was observed with this experiment.
28th experiment is the test of the diffusion term taken as 2, instead of 1. As model documentation didn’t recommend this option, it gave worse results than the default one. Also due to the halt of the run, time step had to be decreased while using this option.
Initialization was at 00z with all experiments until experiment 29. With this experiment, 12z of 14 August 2004, and with 31st experiment, 18z of the same day are taken as initial times. Probably due to the fact that these are more unstable times of the day at the domain area, the scores were worse than those of 00z initializations. With all “good” options, a final experiment, number 30 is performed as a member of the sensitivity analysis and gave the best results though. Table 3.17 shows the RMSE and BIAS statistics for this run. All analysis are done according to this simulation.
Table 3.17: RMSE and BIAS statistics for the second domain of 30th experiment.
parameter RMSE BIAS
T -0.31079173 1.19597328 QVAPOR 0.00001821 0.00070228 QCLOUD 0.00000045 0.00001093 QRAIN 0.00000050 0.00001273 U -0.11205843 1.75780678 V -0.18146902 1.79373264 W -0.00375682 0.05371501 PH -5.92352057 71.40837860 PHB 0.02681876 120.20515442 MUB -0.28473541 180.04959106 U10 0.23694731 1.98098481 V10 -0.81036532 1.97479606
Experiments 31 and 32 were done with ECMWF data as input, instead of GFS data. Although these runs had better results at middle and higher levels of the troposphere, there was a problem with the surface level temperatures and humidity, so that convection was not succesfully created at the boundary layer. To overcome this problem, instead of using the surface fields, interpolation from 1000 hPa level can be tried with next studies.
Furthermore, CHAMP data from COSMIC system was assimilated to create better initial conditions. However, due to the small number of data available, the results did not bring any improvement.
Figure 3.9: Mean sea level pressure difference of the forecast and the observation at 12z on 15.08.2004, according to the sensitivity test elements 15, 16, 17 and 18.
Figure 3.9 (cont.): Mean sea level pressure difference of the forecast and the observation at 12z on 15.08.2004, according to the sensitivity test elements 19, 20, 21, 22, 23 and 24.
Figure 3.9 shows the mean sea level pressure difference between the forecasts of the sensitivity test elements and the observations for the coarse domain, on 15.08.2004 12UTC, when the convection is playing a major role at the area. With a general look, model seems to be overestimating the mean sea level pressure over Mediterranean and lower latitudes. A negative bias can be seen near Caspian Sea, Baltic Sea and
Figure 3.9 (cont.): Mean sea level pressure difference of the forecast and the observation at 12z on 15.08.2004, according to the sensitivity test elements 25, 26, 27, 28.
Figure 3.9 (cont.): Mean sea level pressure difference of the forecast and the observation at 12z on 15.08.2004, according to the sensitivity test element 30.
north part of the Black Sea where the low pressure is effecting. It is obvious that the 23rd run, which was a test for seeing how hydrostatic approach works, has important problems forecasting the mean sea level pressure, especially on mountainous regions like Northern Iran, Anatolia and Alps. The most succesful run according to an overall look at these plots can be analysed as 30th run, which also gave the best numerical results in RMSE and BIAS statistics.
Figure 3.10: 850 hPa level equivalent potential temperature difference of the forecast and the observation at 12z on 15.08.2004, according to the sensitivity test elements 15, 16, 17 and 18.
Figure 3.10 shows the equivalent potential temperature differences of the forecasts and the observations at 850 hPa level, for 15.08.2004 12UTC. It can be commented that the model works fine in general, noting that there are negative and positive bias areas in the domain. The planetary boundary layer scheme test, 19th run features an overestimate over the southeast regions of the domain.
Figure 3.10 (cont.): 850 hPa level equivalent potential temperature difference of the forecast and the observation at 12z on 15.08.2004, according to the sensitivity test elements 19, 20, 21, 22, 23 and 24.
Figure 3.10 (cont.): 850 hPa level equivalent potential temperature difference of the forecast and the observation at 12z on 15.08.2004, according to the sensitivity test elements 25, 26, 27, 28 and 30.
Appendix A.4 is given for a clear look how the parameterization choice effects the forecast with respect to the total precipitation amount, a parameter which can be said to be the effected by all other parameters used in the model. The figures show the fine-scale forecast of precipitation between 05UTC and 06UTC of 15.08.2004, including the rainfall from the severe storm studied. These results can be compared to the TRMM rainfall estimations, given at Appendix A.5.
The most important parameterization is the cumulus parameterization, which is tested by the 15th, 16th, 17th, 24th and 27th runs. It is seen that Grell 3-d ensemble cumulus scheme (16th run) creates additional rainfall over eastern Thracen region and on Marmara Sea, according to Kein-Fritsch scheme. Location of the storm and amount of precipitation from the storm also differs from those of the default scheme. Betts-Miller-Janjic scheme (17th run) can be said to be giving the most different pattern in precipitation prediction. It creates a homogeneous area with mild amounts of rainfall over the region of the storm. Unfortunately, according to the verification data, this is not the case happened.
Result of the Grell-Devenyi ensemble cumulus scheme can be seen on 24th run’s plot, and is also distuingishably different from the other runs. Precipitation seems to be scattered in a wider area over Black Sea, compared to other schemes. Amount of precipitation doesn’t give cell-like higher spots, but can be said to be a better prediction than Betts-Miller-Janjic scheme’s one.
27th run was the test of the Old-Kein-Fritsch scheme. This scheme creates the storm cell shifted a bit northeast from its place in other runs, which is a bit more far away from the real location. Amount of precipitation does not indicate huge difference from the new Kein Fristch scheme.
The precipitation pattern and amount is not only effected by the cumulus parameterization, but also the other physics and dynamics options, as seen from the other plots not mentioned. Since it is the most difficult parameter to forecast, a succesful run can be defined as the one which gives a higher accuracy over this. But verification of the precipitation is also difficult, due to the lack of high resolution observational network. Although Kein-Fritsch was chosen at this study, Grell-Devenyi ensemble and Grell 3-d ensemble schemes are also found succesful.
Additionally, some other forecast-observation difference plots for the “successful” 30th run are included as Figure 3.9, 3.10, 3.11, 3.12, 3.13, 3.14, 3.15 and 3.16.
Figure 3.11 shows the lowest sigma level temperature difference of the forecast and the observation. The forecast over the coasts seem underpredicted, where there is a positive bias over northern inlands.
Figure 3.11 : Lowest sigma level temperature difference of the forecast and the observation at 12z on 15.08.2004, according to the sensitivity test element 30.
At figure 3.12, 1500 m relative vorticity difference is plotted. There seems no problem on forecast of this field, in general. Lateral boundaries create imaginary vorticity patterns.
Figure 3.13 shows the 850 hPa level temperature difference. Forecast of this field seems quite succesful, but there is a slightly negative bias over Alps, Caucasus, northern Iran , northern Algeria and mountaionus regions of Spain.
Figure 3.12 : 1500 m relative vorticity difference of the forecast and the observation at 12z on 15.08.2004, according to the sensitivity test element 30.
Figure 3.13 : 850 hPa level temperature difference of the forecast and the observation at 12z on 15.08.2004, according to the sensitivity test element 30.
Figure 3.14 : 700 hPa level relative humidity difference of the forecast and the observation at 12z on 15.08.2004, according to the sensitivity test element 30.
Figure 3.15 : 500 hPa relative vorticity difference of the forecast and the observation at 12z on 15.08.2004, according to the sensitivity test element 30.
700 hPa level relative humidity difference is plotted on Figure 3.14. The model is capable to forecast the humidity field in general, but there is an important underestimate over Greece.
Figure 3.15 shows the 500 hPa relative vorticity difference between the forecast and the observations. The anomaly occurs near the low center and the cold front.
500 hPa temperature is maybe the most succesful field forecasted by the model. Except a very small area over Alps, the anomaly is smaller than 2 C. Figure 3.16 shows the difference map.
Figure 3.17 shows the 925 hPa equivalent potential temperature difference. There are more errors at this level when compared to that of 850 hPa, because of land-atmosphere interactions and boundary layer effects. There seems an underestimate over seas and positive bias over lands near the coasts.
Figure 3.16 : 500 hPa temperature difference of the forecast and the observation at 12z on 15.08.2004, according to the sensitivity test element 30.
Figure 3.17 : 925 hPa equivalent potential temperature difference of the forecast and the observation at 12z on 15.08.2004, according to the sensitivity test element 30.
Figure 3.18 : CAPE difference of the forecast and the observation at 12z on 15.08.2004, according to the sensitivity test element 30.
At last, CAPE difference is plotted and shown as Figure 3.18. There seems important biases of this field. When looked at the study area, Marmara Region, it can be seen that there is an overestimate over Black Sea, and underestimate over Southern and Eastern Marmara. This is the place where severe convection occurred. This plot shows that the model is able to forecast the severe convection, but the location of the forecast is shifted to Black Sea. Though the real tornado and supercell occurred over Yalova, where model predicts the cell over a northern part of Istanbul.
As a conclusion to sensitivity analysis, Kein-Fritsch scheme is used as cumulus parameterization, noting the two Grell schemes are also succesful. WRF Single-Moment 3-class scheme is chosen as microphysics parameterization instead of Eta microphysics. Shortwave radiation scheme Dudhia, and longwave radiation scheme RRTM were preferred instead of Eta radiation schemes. Surface physics is changed into Noah Land Surface Model with 4 layers, from 5-layer thermal diffusion. Surface layer parameterization is switched into Eta similarity scheme, from the default MM5 similarity scheme. For planetary boundary layer, again Eta similarity (Mellor-Yamada-Janjic scheme) found to be better than the default Yonsei University scheme. 6th-order horizontal hyper diffusion option is used as 2, relaxation at the boundaries were done with 4 points, and the run is performed nonhydrostatically with full diffusion option, instead of simple diffusion.
The horizontal resolution of the first domain was 24 km, and the inner domains had 8 and 2.67 km resolution, with 1:3:3 aspect ratio. Number of vertical levels for each domain was 45. Time step is chosen as 72 seconds, and for the inner domains it became 24 seconds and 8 seconds respectively. Runs are performed with 14.08.2004 00UTC and 15.08.2004 00UTC initialization.
Before coming to the meteorological analysis of the case, the namelist variables of 30th run will be given here.
The namelist.wps file is as follows: &share
wrf_core = 'ARW', max_dom = 3,
end_date = '2004-08-16_00:00:00','2004-08-16_00:00:00','2004-08-16_00:00:00' interval_seconds = 21600 io_form_geogrid = 2, / &geogrid parent_id = 1, 1, 2, parent_grid_ratio = 1, 3, 3, i_parent_start = 1, 60, 202, j_parent_start = 1, 20, 115, e_we = 220, 340, 163, e_sn = 170, 340, 121, geog_data_res = '10m','2m','30s' dx = 24000, dy = 24000, map_proj = 'lambert', ref_lat = 45.00, ref_lon = 20.00, truelat1 = 35.0, truelat2 = 50.0, stand_lon = 30.0, geog_data_path = '/home/kahraman/WRF/geog' / &ungrib out_format = 'WPS', prefix = 'FILE', / &metgrid fg_name = 'FILE' io_form_metgrid = 2, / &mod_levs press_pa = 201300 , 200100 , 100000 , 95000 , 90000 , 85000 , 80000 , 75000 , 70000 , 65000 , 60000 , 55000 , 50000 , 45000 , 40000 , 35000 , 30000 , 25000 , 20000 , 15000 , 10000 , 5000 , 1000 /
